No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
A novel scalable method for machine degradation assessment using deep convolutional neural network
Abstract
Bandsaw machines are widely used in the rough machining stage to cut various materials into required dimensions. Deterioration on the blade, which is a critical component of the bandsaw machine, not only causes a waste of cutting material but also represents a major portion of the operation & maintenance cost for the machine user. Although non-high-end manufactures put as much emphasis on the accuracy of the cuts as high-end manufacturers, non-high-end bandsaw machine users are not as easily able to justify the high cost associated with the blade wear monitoring solution. Therefore, this paper proposes a methodology to develop a scalable blade degradation model that is suitable for massive deployment at an affordable cost. A 4-stage
roadmap is proposed to provide step by step guidance in the development and deployment of the scalable blade degradation model. As the core issue of the roadmap, the degradation model
development is solved by the proposed dual-phase modeling methodology. In phase I, a physics informed model (which relies on physical analysis to extract effective features) is established to
generate a reliable health indicator (HI) to monitor the blade wear condition utilizing the critical vibration and acoustic signals. Phase II proposes to develop a deep convolutional neural network (DCNN) based surrogate model to replace the physics informed model. The DCNN based surrogate model will use only alternative low-cost sensor data. By eliminating the usage of the high-cost vibration and acoustic sensors, the developed surrogate model is expected to cost much
less than the physics informed model. Finally, the effectiveness of the proposed methodology is validated using data from different bandsaw machines and blades, and the superior performance of the DCNN is observed as compared to traditional machine learning algorithms.
... Compared to the autoencoders and DBNs, more research has been conducted on CNNs for intelligent machining, focusing on tool wear monitoring [89][90][91][92][93][94][95][96][97][98][99][100][101][102][103][104][105][106][107]. It has also been used for chatter detection [108,109] and surface roughness prediction [110]. ...
... During the machining process, the acquired sensory data can be combined with the experience-and physics-based features from the process to build low-cost models for process monitoring. Li et al. [104] developed a model, which relies on physical analysis to extract useful features to establish a reliable health indicator for tool condition monitoring utilizing the vibration and acoustic signals. Then, they developed a deep CNN model using 20 low-cost processes and cut variables to replace the physicsbased model (Fig. 13). ...
... The model establishes a health index using the statistical-, experience-, and physics-based features. The established index is then considered the target of the CNN model, which was trained using lowcost sensory data ( [104] with permission from Elsevier) feature matrix for wear prediction in the milling process using CNN. Thus, for each sample data, a total of 54 multi-domain features were extracted to form a column of the original feature matrix. ...
Data-driven methods provided smart manufacturing with unprecedented opportunities to facilitate the transition toward Industry
4.0–based production. Machine learning and deep learning play a critical role in developing intelligent systems for descriptive,
diagnostic, and predictive analytics for machine tools and process health monitoring. This paper reviews the opportunities and
challenges of deep learning (DL) for intelligent machining and tool monitoring. The components of an intelligent monitoring
framework are introduced. The main advantages and disadvantages of machine learning (ML) models are presented and compared with those of deep models. The main DL models, including autoencoders, deep belief networks, convolutional neural
networks (CNNs), and recurrent neural networks (RNNs), were discussed, and their applications in intelligent machining and tool
condition monitoring were reviewed. The opportunities of data-driven smart manufacturing approach applied to intelligent
machining were discussed to be (1) automated feature engineering, (2) handling big data, (3) handling high-dimensional data,
(4) avoiding sensor redundancy, (5) optimal sensor fusion, and (6) offering hybrid intelligent models. Finally, the data-driven
challenges in smart manufacturing, including the challenges associated with the data size, data nature, model selection, and
process uncertainty, were discussed, and the research gaps were outlined.
... In the literature, the existing methods for tool wear monitoring can be categorized as direct methods and indirect methods [2]. The indirect methods evaluate the tool wearing based on the real-time acoustic and vibration signals [3], spindle load profiles [4], temperature [5] etc. Various forces can also monitor tool wear [6,7]. Although many successful applications are reported in the literature, the reliability of the indirect method is difficult to be calibrated and the performance varies case by case. ...
... In [3], vibration and acoustic signal is used to establish a physic model to monitor the blade wear condition and generate health indicator, then an effective deep convolutional neural network (DCNN) replace the physic model to reduce the cost. In [6], an experimental and analytical method based on the measurement of cutting forces (static and dynamic) and vibration is proposed. ...
... P.Li et. al. [3] explores the degradation of band saw blades based on the vibration and acoustic sensors. X. Jia [4] studies the degradation of boring tools by using the spindle load profile that is collected from the controller. ...
In Liquid Crystal Display (LCD) panel cutting, in-line monitoring of tool wear is important to maintain the dimension precision of finished products and to avoid possible damages to the workpiece. However, limited by the space for camera installation and the miniature structure of the tool itself, monitoring the wear of LCD panel cutting wheel is not a simple task. In this research, we proposed a machine-vision based instrumentation system and a systematic methodology for cutting wheel degradation monitoring. The proposed method describes the degradation of cutting wheel by estimating the tooth height of the cutting wheel blade based on the partially observed random samples. A series of methods are proposed to improve the reliability of the results. The effectiveness of the proposed method is validated based on the field data that is collected from three maintenance cycles. The validation results demonstrate consistent degradation trend of the cutting wheel over production cycles.
... Reduces errors and uncertainty [72,78] [ 71,77] Physics-based models may perform lower in some applications [15] Makes the ML model physically meaningful [13,28] Can detect and localize faults with high accuracy Reduced requirement of training data ...
... Li et al.[77] proposed a PBML method for the estimation of the blade wear of the bandsaw machines. Vibration and acoustic signals are used for the surrogate physicsinformed model. ...
The improvements in wind energy infrastructure have been a constant process throughout many decades. There are new advancements in technology that can further contribute towards the Prognostics and Health Management (PHM) in this industry. These advancements are driven by the need to fully explore the impact of uncertainty, quality and quantity of data, physics-based machine learning (PBML), and digital twin (DT). All these aspects need to be taken into consideration to perform an effective PHM of wind energy infrastructure. To address these aspects, four research questions were formulated. What is the role of uncertainty in machine learning (ML) in diagnostics and prognostics? What is the role of data augmentation and quality of data for ML? What is the role of PBML? What is the role of the DT in diagnostics and prognostics? The methodology used was Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA). A total of 143 records, from the last five years, were analyzed. Each of the four questions was answered by discussion of literature, definitions, critical aspects, benefits and challenges, the role of aspect in PHM of wind energy infrastructure systems, and conclusion.
... Lee et al. [110] proposed a methodology to use an Elman neural network model to predict long-term performance of bridge elements. Other recent studies [111][112][113][114][115] proposed degradation models of bridge elements using machine learning and artificial intelligence techniques. In addition, sewer pipes condition prediction models based on artificial intelligence techniques can be found in Ref. [116]. ...
... Other models [117][118][119] Two-phase degradation Rolling element bearing [120] Two-phase Wiener and gamma Liquid coupling diode [121] Multiphase Wiener High storage capacitor [77][78][79] Time series model Rotating machinery [76,80] Long memory models Long-term time series data [106] Neural network model Rolling contact thrust bearings [108] Neural network model Highway culverts [109] Recurrent neural network model and support vector classifier Pavement performance [110] Elman neural network model Bridge elements [107,[111][112][113][114][115][116] Various machine learning techniques Bridges, water and sewer pipes ...
Safety and reliability of large critical infrastructure such as long-span bridges, high-rise buildings, nuclear power plants, high-voltage transmission towers, rotating machinery, and so on, are important for a modern society. Research on reliability and safety analysis started with a “small data” problem dealing with relative scarce lifetime or failure data. Later, degradation modeling that uses performance deterioration, or, condition data collected from in-service inspections or online health monitoring became an important tool for reliability prediction and maintenance planning of highly reliable engineering systems. Over the past decades, a large number of degradation models have been developed to characterize and quantify the underlying degradation mechanism using direct and indirect measurements. Recent advancements in artificial intelligence, remote sensing, big data analytics, and Internet of things are making far-reaching impacts on almost every aspect of our lives. The effect of these changes on the degradation modeling, prognosis, and safety management is interesting questions to explore. This paper presents a comprehensive, forward-looking review of the various degradation models and their practical applications to damage prognosis and management of critical infrastructure. The degradation models are classified into four categories: physics-based, knowledge-based, data-driven, and hybrid approaches.
... Embeded Tacit [195], [196], [170], [58], [197], [188], [13], [53], [55], [198], [174], [144] ✓ Simulator [199], [59], [171], ✓ Gauge [199], [59], [171], [200], [183], [167], [201], [169], [169] ✓ Extractor [182], [36], [62], [172,213,63], [113], [149] ✓ Operator [118], [43], [165], [51] ✓ Operator [81], [130], [209] ✓ Operator [37], [210], [113], [217] ✓ Structure blueprint [114], [129], [121], [115], [186], [179], [50], [120], [91], [114] [156], [178] [95], [166], [211], [212], [20], [107] ✓ Structure blueprint [57], [142], [76], [149], [187] ✓ Initializer [154], [168] ✓ Initializer [96] , [133], [123] [132], [150] ✓ Consistency [177], [214] ✓ Consistency [146], [131] ✓ Consistency [176], [84] ✓ Conflict [143], [126] , [173] ✓ Conflict 1. Due to explicit analytical equations or models that define clear input-output mathematical relationships, explicit knowledge is the most common way for building PIML. It is widely used in the construction of "simulators", "extractors", "operators", and "consistency checks". ...
... A new intrusion detection approach using FHD-CNN was proposed in [27], highlighting the limitations of traditional systems and the need for carefully curated training data and constant updates to adapt to evolving threats. A method for assessing machine degradation was suggested in [28] using DCNN and transfer learning techniques. In [20], To increase accuracy, an optimized ANN-based technique was developed by investigating alternative hyperparameters, applying ensemble methods, and using larger and diverse datasets. ...
Real-time deep learning faces the challenge of balancing accuracy and time, especially in cybersecurity where intrusion detection is crucial. Traditional deep learning techniques have been insufficient in identifying network anomalies and intrusions. To address this, a Fully Streaming Big Data Framework based on optimized Deep Learning for cybersecurity (FSBDL) was proposed. The framework leverages two parallel optimization algorithms, Adam and RMSprop, labeled Hyper-parallel optimization (HPO) techniques to enhance efficiency and stability. The optimized CNN in the proposed framework achieves high accuracy in real-time intrusion detection without compromising reliability. The CNN is customized to address overfitting issues in recurrent connections by reducing the number of training parameters, using customized activation functions and regularization techniques. The CNN is trained in parallel using Adam and RMSprop optimization algorithms, resulting in significant accuracy improvements that surpass traditional methods and current state-of-the-art approaches. The HPO is a crucial component of the proposed framework, as it enables the system to detect intrusions in real-time, ensuring prompt response to potential cyber threats. The six-layer FSBDL framework includes data pre-processing, feature selection, hyper-parallelism, a customized CNN, a GUI layer for interpretation, and a detection-evaluation layer. The optimized CNN was designed to detect intrusions in real-time without compromising accuracy or reliability. The proposed algorithms were evaluated using various performance metrics, showing that the accuracy of the framework surpasses 99.9%, indicating its superiority over other intrusion detection models. This novel intrusion detection model offers promising prospects for cybersecurity, and its effectiveness and accuracy have been demonstrated through experimentation.
... HI quality is an important factor affecting the RUL prediction accuracy; another essential factor is the learning capability of intelligent computational algorithms. With the improvement in computing power, deep learning (DL) [11], including long short-term memory (LSTM) [12,13], deep belief network (DBN) [14], and CNN [15][16][17], has become a hot research topic. Ding et al. [18] considered the transition of different degradation stages and proposed one-and multi-stage iteration prediction models based on the LSTM neural network for RUL prediction. ...
As a critical content of condition-based maintenance (CBM) for mechanical systems, remaining useful life (RUL) prediction of rolling bearing attracts extensive attention to this day. Through mining the bearing degradation rule from operating data, the deep learning method is often used to perform RUL prediction. However, due to the complexity of operating data, it is usually difficult to establish a satisfactory deep learning model for accurate RUL prediction. Thus, a novel convolutional neural network (CNN) prediction method based on similarity feature fusion is proposed. In this paper, the similarity features are extracted based on the correlation between statistical features and time series. After sensitive feature screening, eligible features are applied to develop a health indicator (HI), which can be used to define the bearing failure stages and reduces the complexity of the CNN model. Subsequently, a one-dimensional CNN is established to predict the RUL of bearing, and the HI is utilized to train the prediction model. The proposed approach is verified by FEMTO bearing datasets and IMS bearing datasets. And the experimental results reveal the superiority and effectiveness of the feature fusion-based CNN method in constructing HI and accurate RUL prediction.
... In the field of rotating machinery component failure testing, as described in this paper, intelligent fault diagnosis appears to have become more widespread among researchers [12][13][14][15][16][17][18]. It is also broadly discussed in review works. ...
Linear and nonlinear analyses served to detect the degree of tooth damage in a cylindrical gear. The gear in question is widely used in industrial applications and is often subject to gear tooth damage. Multiple-sourced vibrations affecting the test stand may distort or conceal the vibration components responsible for tooth failure. Given its construction, the test stand offers low stiffness, and additional control and measurement equipment further contribute to overall vibrations. Therefore, the vibration signal is likely to contain various disturbances, which further hinder correct diagnosis. Basic vibroacoustic signal analysis methods, which were employed at a preliminary stage, failed to produce unambiguous and satisfactory results. Hence, nonlinear recurrence analysis of gear operation dynamics was applied. It is assumed that the occurrence of damage to the gear will impede the regularity of drive transmission. The adopted research method addresses the limitations of a two-state diagnosis of gear condition (no damage/damage) by allowing the extent of gear tooth damage to be specified.
... Through layer-by-layer convolution operation, CNN can extract the spatial features hidden inside the data [31]. The features can be further applied for classification or regression. ...
Accurate degradation tendency prediction (DTP) is vital for the secure operation of a pumped storage unit (PSU). However, the existing techniques and methodologies for DTP still face challenges, such as a lack of appropriate degradation indicators, insufficient accuracy, and poor capability to track the data fluctuation. In this paper, a hybrid model is proposed for the degradation tendency prediction of a PSU, which combines the integrated degradation index (IDI) construction and convolutional neural network-long short-term memory (CNN-LSTM). Firstly, the health model of a PSU is constructed with Gaussian process regression (GPR) and the condition parameters of active power, working head, and guide vane opening. Subsequently, for comprehensively quantifying the degradation level of PSU, an IDI is developed using entropy weight (EW) theory. Finally, combining the local feature extraction of the CNN with the time series representation of LSTM, the CNN-LSTM model is constructed to realize DTP. To validate the effectiveness of the proposed model, the monitoring data collected from a PSU in China is taken as case studies. The root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE) obtained by the proposed model are 1.1588, 0.8994, 0.0918, and 0.9713, which can meet the engineering application requirements. The experimental results show that the proposed model outperforms other comparison models.
... Convolutional Neural Networks (CNN), which were originally designed for image classification purposes [13], are now widely used for analyzing complex and fluctuating vibration signals of bearings, induction motors, gearboxes etc. [14] because of their significant characteristics such as shared network weights, locally receptive fields and spatial subsampling (spatial pooling) [15] [16][17] [18]. Using multiple filters, the convolutional layers of the network perform convolutional operation on the raw input data in order to generate meaningful feature maps. ...
Intelligent data-driven machinery health identification has been attracting increasing attention in the manufacturing industries, due to reduced maintenance cost and enhanced operation safety. Despite the successful development, the main limitation of most existing methods lies in the assumption that the training and testing data are collected from the same distribution, i.e. the same machine under identical condition. However, this assumption is difficult to be met in the real industries, since the diagnostic model is generally expected to be applied on new machines. In order to address this issue, a deep learning-based cross-machine health identification method is proposed for industrial vacuum pumps, which are of great importance in the manufacturing industry but have received far less research attention in the literature. Generalized diagnostic features can be learnt using the proposed domain adaptation technique with maximum mean discrepancy metric. The health identification model learnt from the training machines can be well applied on new machines. Experiments on a real-world vacuum pump dataset validate the proposed method, which is promising for industrial applications.
... For industry use, Wang et al. [50] presented an extensive survey and feasibility by comparing traditional machine learning to deep learning in manufacturing application, reflecting a growing trend in implementing integrated intelligent manufacturing systems to leverage human operation [51]. On the other hand, deep learning approaches can be as well applied to tool wear assessment modeling [52], indicating another possibility of monitoring equipment status for manufacturers. ...
Quality and efficiency are crucial indicators of any manufacturing company. Many companies are suffering from a shortage of experienced workers across the production line to perform complex assembly tasks. To reduce time and error in an assembly task, a worker-centered system consisting of multi-modal Augmented Reality (AR) instructions with the support of a deep learning network for tool detection is introduced. The integrated AR is designed to provide on-site instructions including various visual renderings with a fine-tuned Region-based Convolutional Neural Network, which is trained on a synthetic tool dataset. The dataset is generated using CAD models of tools and displayed onto a 2D scene without using real tool images. By experimenting the system to a mechanical assembly of a CNC carving machine, the result of a designed experiment shows that the system helps reduce the time and errors of the given assembly tasks by 33.2 % and 32.4 %, respectively. With the integrated system, an efficient, customizable smart AR instruction system capable of sensing, characterizing requirements, and enhancing worker’s performance has been built and demonstrated.
... Then an enhanced convolutional LSTM network was designed for damage monitoring of the automotive suspension component. Li et al. [193] developed a scalable degradation assessment approach for bandsaw machine by proposing a dual-phase modeling method. In this approach, a physics informed model is firstly established to generate the HI to monitor wear condition using the vibration and acoustic signals. ...
With the rapid development of manufacturing industry, machine fault diagnosis has become increasingly significant to ensure safe equipment operation and production. Consequently, multifarious approaches have been explored and developed in the past years, of which intelligent algorithms develop particularly rapidly. Convolutional neural network, as a typical representative of intelligent diagnostic models, has been extensively studied and applied in recent five years, and a large amount of literature has been published in academic journals and conference proceedings. However, there has not been a systematic review to cover these studies and make a prospect for the further research. To fill in this gap, this work attempts to review and summarize the development of the Convolutional Network based Fault Diagnosis (CNFD) approaches comprehensively. Generally, a typical CNFD framework is composed of the following steps, namely, data collection, model construction, and feature learning and decision making, thus this paper is organized by following this stream. Firstly, data collection process is described, in which several popular datasets are introduced. Then, the fundamental theory from the basic convolutional neural network to its variants is elaborated. After that, the applications of CNFD are reviewed in terms of three mainstream directions, i.e. classification, prediction and transfer diagnosis. Finally, conclusions and prospects are presented to point out the characteristics of current development, facing challenges and future trends. Last but not least, it is expected that this work would provide convenience and inspire further exploration for researchers in this field.
In this study, we describe the development of a Deep Convolutional Neural Network (DCNN)-based system that gives warnings about soon to be necessary blade changes of log bandsaw machines. The major difficulties in monitoring the blade wear arise from the complex domain of primary wood machining and a challenging small dataset. The developed concatenated DCNN can accurately indicate necessary saw blade changes, whereas purely feature-based and Multi-Layer-Perceptron-based methods fail. The system can be applied directly without complex preprocessing. To understand the effects leading to the DCNN’s success, we also present an analysis of the prediction behavior in comparison to classical approaches.
In this work, both hardware and software modifications in a typical research reactor protection system (RPS) is proposed. The reactor cooling pumps are tripped based on vibrations safety signals of the pumps while the reactor SCRAM signal is initiated based on low flow rate and pressure drop across the reactor core which is a direct result of pumps trip. The main objective of this work is to develop reactor SCRAM signal based of core cooling pumps vibration signals. The early shutdown of the reactor based on pumps vibration signals is of significant importance not only in cooling the decay power of the reactor core after shutdown but also to prevent pumps failure. In the hardware model, the core cooling pumps vibration signals are feed to RPS to initiate reactor SCRAM signal. In the software model, a modular artificial neural network (ANN) is used in modeling the vibration monitoring of the research reactor (ETRR-2). The input and the output signals of the vibration transducer are used as a source data for training the neural network model. The type of the network used in this methodology is the supervised Multilayer Feed-Forward Neural Networks with the back-propagation (BP) algorithm. Vibration analysis programs are used in research reactors (RRs) to identify faults in machinery, plan machinery repairs, and keep machinery functioning for as long as possible without failure. The vibration severity limits are determined based on the International Organization for Standardization (ISO) 10816. The ANNs were designed using two different methods; one is by using hardware application contained two out of three voting and dynamic modules for trip signal by using ANNs. The current model classifies the vibration signals into five ranges low, good, satisfactory, unsatisfactory, and unacceptable vibration. The ANN is trained to detect the signal and vote to take the correct and safe action. The results demonstrate that the ANN can help in taking predictive actions for the safe core coolant pumps operation.
Given the growing amount of industrial data in the 4th industrial revolution, deep learning solutions have become popular for predictive maintenance (PdM) tasks, which involve monitoring assets to anticipate their requirements and optimise maintenance tasks. However, given the large variety of such tasks in the literature, choosing the most suitable architecture for each use case is difficult. This work aims to facilitate this task by reviewing various state-of-the-art deep learning (DL) architectures and analysing how well they integrate with predictive maintenance stages to meet industrial companies’ requirements from a PdM perspective. This review includes a self-organising map (SOM), one-class neural network (OC-NN) and generative techniques. This article explains how to adapt DL architectures to facilitate data variability handling, model adaptability and ensemble learning, all of which are characteristics relevant to industrial requirements. In addition, this review compares the results of state-of-the-art DL architectures on a publicly available dataset to facilitate reproducibility and replicability, enabling comparisons. Furthermore, this work covers the mitigation step with deep learning models, the final PdM stage that is essential for implementing PdM systems. Moreover, state-of-the-art deep learning architectures are categorised, analysed and compared; their industrial applications are presented; and an explanation of how to combine different architectures in a solution is presented that addresses their gaps. Finally, open challenges and possible future research paths are presented and supported in this review, and current research trends are identified.
Monitoring the degradation of machines to anticipate potential failures represents a significant challenge. In Industry 4.0, this task is critical when the costs associated with maintenance and stoppages on the productive processes are high. Nowadays, many preventive maintenance techniques employ supervised or unsupervised machine learning algorithms. However, the definition of which features should be processed by such algorithms is not a simple task, being crucial to the proposed technique’s success. Against this background, we consider whether unsupervised algorithms combined with time-series decomposition can enhance the estimate of a machine’s health. This article proposes HealthMon as a novel approach to compute a health index of machines based on sensor measurements. HealthMon extracts time-series from such sensors, which are decomposed in an unsupervised way to present the health state along time. The health index is related to the degradation of the considered machine, thus optimizing the machine maintenance schedule. This work advances the state-of-the-art in the following ways: (i) it proposes a novel index of machines health, which yields a more direct and intuitive view of machine degradation; (ii) it devises the first approach capable of estimating the health index of a machine in a completely unsupervised way; (iii) it generalizes vibrating and rotating machines, thus being able to monitor a wide range of industrial equipment. We evaluated our method using both simulated and real data. The results show that the evolution of vibrating machines’ failures can be effectively detected under various input workloads. Finally, through HealthMon, industry decision-makers benefit from the guidelines for preventive actions at appropriate times, thus meeting Industry 4.0.
This study aims to present an overall review of the recent research status regarding Machine Learning (ML) applications in machining processes. In the current industrial systems, processes require the capacity to adapt to manufacturing conditions continuously, guaranteeing high performance- in terms of production quality and equipment availability. Artificial Intelligence (AI) offers new opportunities to develop and integrate innovative solutions in conventional machine tools to reduce undesirable effects during operational activities. In particular, the significant increase of the computational capacity may permit the application of complex algorithms to big data volumes in a short time, expanding the potentialities of ML techniques. ML applications are present in several contexts of machining processes, from roughness quality prediction to tool condition monitoring. This review focuses on recent applications and implications, classifying the main problems that may be solved using ML related to the machining quality, energy consumption and conditional monitoring. Finally, a discussion on the advantages and limits of ML algorithms is summarized for future investigations.
Given the growing amount of industrial data spaces worldwide, deep learning solutions have become popular for predictive maintenance, which monitor assets to optimise maintenance tasks. Choosing the most suitable architecture for each use-case is complex given the number of examples found in literature. This work aims at facilitating this task by reviewing state-of-the-art deep learning architectures, and how they integrate with predictive maintenance stages to meet industrial companies' requirements (i.e. anomaly detection, root cause analysis, remaining useful life estimation). They are categorised and compared in industrial applications, explaining how to fill their gaps. Finally, open challenges and future research paths are presented.
With the rapid development of manufacturing industry, machine fault diagnosis has become increasingly significant to ensure safe equipment operation and production. Consequently, multifarious approaches have been explored and developed in the past years, of which intelligent algorithms develop particularly rapidly. Convolutional neural network (CNN), as a typical representative of intelligent diagnostic models, has been extensively studied and applied in recent five years, and a large amount of literature has been published in academic journals and conference proceedings. However, there has not been a systematic review to cover these studies and make a prospect for the further research. To fill in this gap, this work attempts to review and summarize the development of the Convolutional Network based Fault Diagnosis (CNFD) approaches comprehensively. Generally, a typical CNFD framework is composed of the following steps, namely, data collection, model construction, and feature learning and decision making, thus this paper is organized by following this stream. Firstly, data collection process is described, in which several popular datasets are introduced. Then, the fundamental theory from the basic CNN to its variants is elaborated. After that, the applications of CNFD are reviewed in terms of three mainstream directions, i.e. classification, prediction and transfer diagnosis. Finally, conclusions and prospects are presented to point out the characteristics of current development, facing challenges and future trends. Last but not least, it is expected that this work would provide convenience and inspire further exploration for researchers in this field.
Remaining useful life (RUL) prediction has been a difficult problem in industry and academia. Deep learning techniques based on supervised learning have made great progress compared to traditional prediction algorithms. However, due to the random nature of system degradation behavior, deep supervised learning cannot accurately adapt the model obtained based on the training set to the systems in the testing set. The similarity-based prediction method has achieved accurate approximate remaining life, which called similar RUL. Therefore, the combination of supervised learning and similarity-based prediction has the potential to take advantages of the learning ability of the network model and similar RUL to improve the prediction accuracy. This paper investigates the effect of using similarity to improve the prediction accuracy of deep supervised learning. In addition, since health indicator (HI) and target labels are the essential factors of supervised learning, this paper proposes a method for constructing HI based on unsupervised learning. Then according to the constructed HI, a difference-based method proposed to generate target labels. To verify the effectiveness of the proposed method, this paper conducted experimental verification on the C-MAPSS engine dataset and XJTU-SY bearing dataset and made related comparative experiments. Experimental results show that the proposed method obtains favorable results against the traditional methods.
In this paper, a methodology for assessing the unpredictability of systems with memory was developed. The proposed approach consists in approximating the probability distribution exhibited by the response of a system, understood as a stochastic process, with a deep recurrent neural network; such networks offer increased forecasting capability by exploiting an accumulative register of previous system states. Once the probability distribution is computed, the uncertainty or entropy of the underlying process is measured. This measure determines the degree of regularity in the system, and identifies how atypical the system dynamics are. The proposed model was validated by identifying industrial gas turbine engine faults from recorded sensor data.
As more and more data become available for machine prognostic analysis in the big data environment, effective data suitability assessment methods become highly desired to help locate data with sufficient quality for analysis. Driven by this purpose, this paper proposes a novel and systematic methodology for data suitability assessment based on the needs of prognostics and health management (PHM). In this study, the data suitability for PHM is assessed from the aspects of detectability, diagnosability and trendability, which correspond to the three major tasks of PHM -- fault detection, fault diagnosis, and degradation assessment. The proposed methodology is mainly built upon the recent researches on maximum mean discrepancy in the field of machine learning, which include a family of test statistics that are used to test the difference between two data distributions. The effectiveness of the proposed methodology is demonstrated in diverse industrial applications, which include semiconductors, boring tool degradation and sensorless drive diagnosis. The results in the case studies indicate that the proposed methodology can be a promising tool to evaluate whether the data under study or the extracted feature set is suitable for PHM tasks.
Successful applications of Diffusion Map (DM) in machine failure detection and diagnosis
have been reported in several recent studies. DM provides an efficient way to visualize
the high-dimensional, complex and nonlinear machine data, and thus suggests more
knowledge about the machine under monitoring. In this paper, a DM based methodology
named as DM-EVD is proposed for machine degradation assessment, abnormality detection
and diagnosis in an online fashion. Several limitations and challenges of using DM
for machine health monitoring have been analyzed and addressed. Based on the proposed
DM-EVD, a deviation based methodology is then proposed to include more dimension
reduction methods. In this work, the incorporation of Laplacian Eigen-map and Principal
Component Analysis (PCA) are explored, and the latter algorithm is named as PCA-Dev
and is validated in the case study. To show the successful application of the proposed
methodology, case studies from diverse fields are presented and investigated in this work.
Improved results are reported by benchmarking with other machine learning algorithms
The effect of varying the number of saw blade teeth while transversally cutting beech (Fagus sylvatica L.) wood on the noise level and saw blade lifetime between two sharpenings was tested. The experiment was carried out with raw beech wood samples with dimensions of 25 x 100 x 1000 mm and circular saw blades with cemented carbide tips (24, 40, and 60 teeth). The saw blade diameters were identical (D = 250 mm), as were the cutting wedge angle geometries (α = 15°, β = 60°, γ = 15°). The saw blades were selected based on commonly used blades (in the Czech Republic and Slovakia) for the transversal cutting of the given wood species. Neither the cutting speed (vc = 62 m/s) nor the feed force (Fp = 75 N) were changed during the cutting process. The results suggest that the number of saw blade teeth is an important factor that affects the noise level of saw blade during sawing as well as the wear of cutting edge.
This paper presents a health assessment methodology, as well as specific residual processing and figure of merit algorithms for anemometers in two different configurations. The methodology and algorithms are applied to data sets provided by the Prognostics and Health Management Society 2011 Data Challenge. The two configurations consist of the "paired" data set in which two anemometers are positioned at the same height, and the "shear" data set which includes an array of anemometers at different heights. Various wind speed statistics, wind direction, and ambient temperature information are provided, in which the objective is to classify the anemometer health status during a set of samples from a 5 day period. The proposed health assessment methodology consists of a set of data processing steps that include: data filtering and pre-processing, a residual or difference calculation, and a k-means clustering based figure of merit calculation. The residual processing for the paired data set was performed using a straightforward difference calculation, while the shear data set utilized an additional set of algorithm processing steps to calculate a weighted residual value for each anemometer. The residual processing algorithm for the shear data set used a set of auto-associative neural network models to learn the underlying correlation relationship between the anemometer sensors and to calculate a weighted residual value for each of the anemometer wind speed measurements. A figure of merit value based on the mean value of the smaller of the two clusters for the wind speed residual is used to determine the health status of each anemometer. Overall, the proposed methodology and algorithms show promise, in that the results from this approach resulted in the top score for the PHM 2011 Data Challenge Competition. Using different clustering algorithms or density estimation methods for the figure of merit calculation is being considered for future work.
The 2010 PHM data challenge focuses on the remaining useful life (RUL) estimation for cutters of a high speed CNC milling machine using measurements from dynamometer, accelerometer, and acoustic emission sensors. We present a multiple model approach for wear depth estimation of milling machine cutters using the provided data. The feature selection, initial wear estimation and multiple model fusion components of the proposed algorithm are explained in details and compared with several alternative methods using the training data. The final submission ranked #2 among professional and student participants and the method is applicable to other data driven PHM problems.
Advances in sensor technology have led to an increased interest in using degradation-based sensory information to predict the remaining lives of partially degraded components and systems. This paper presents a stochastic degradation modeling framework for computing and continuously updating remaining life distributions (RLDs) using in situ degradation signals acquired from individual components during their operational lives. Unfortunately, these sensory-updated RLDs cannot be characterized using parametric distributions and their moments do not exist. Such difficulties hinder the implementation of this sensor-based framework, especially from the standpoint of computational efficiency of embedded algorithms. In this paper, we identify an approximate procedure by which we can compute a conservative mean of the sensory-updated RLDs and express the mean and variance using closed-form expressions that are easy to evaluate. To accomplish this, we use the first passage time of Brownian motion with positive drift, which follows an inverse Gaussian distribution, as an approximation of the remaining life. We then show that the mean of the inverse Gaussian is a conservative lower bound of the mean remaining life using Jensen's inequality. The results are validated using real-world vibration-based degradation information. [DOI: 10.1115/1.3159045]
The bandsawing as a multi-point cutting operation is the preferred method for cutting off raw materials in industry. Although cutting off with bandsaw is very old process, research efforts are very limited compared to the other cutting process. Appropriate online tool condition monitoring system is essential for sophisticated and automated machine tools to achieve better tool management. Tool wear monitoring models using artificial neural network are developed to predict the tool wear during cutting off the raw materials (American Iron and Steel Institute 1020, 1040 and 4140) by bandsaw. Based on a continuous data acquisition of cutting force signals, it is possible to estimate or to classify certain wear parameters by means of neural networks thanks to reasonably quick data-processing capability. The multi-layered feed forward artificial neural network (ANN) system of a 6 × 9 × 1 structure based on cutting forces was trained using error back-propagation training algorithm to estimate tool wear in bandsawing. The data used for the training and checking of the network were derived from the experiments according to the principles of Taguchi design of experiments planned as L
27. The factors considered as input in the experiment were the feed rate, the cutting speed, the engagement length and material hardness. 3D surface plots are generated using ANN model to study the interaction effects of cutting conditions on sawblade. The analysis shows that cutting length, hardness and cutting speed have significant effect on tooth wear, respectively, while feed rate has less effect. In this study, the details of experimentation and ANN application to predict tooth wear have been presented. The system shows that there is close match between the flank wear estimated and measured directly.
Gearbox is a critical transmission component in the drivetrain of wind turbine having a dominant failure rate and a highest downtime loss among all wind turbine subsystems. Ensemble empirical mode decomposition and principal component analysis have been extensively investigated for signal decomposition and fault feature extraction from the signals of a wind turbine gearbox. However, presence of background noise in wind turbine signals restricts the applicability of these algorithms in real scenarios. To solve this problem, a novel performance degradation assessment method based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and kernel principal component analysis (KPCA) was proposed to de-noise and fuse vibration signals. A comparison is conducted between CEEMDAN-KPCA and other five cross combinations. After feature extraction, extreme learning machine (ELM) optimized by fruit fly of algorithm (FOA) is employed to predict the remaining use life (RUL) of wind turbine gearbox. To avoid the tedious hand-crafting hidden nodes, a dynamic adjustment strategy is presented for the improvement of search efficiency. The effectiveness of the proposed method is validated using simulated and experimental vibration signals. The results illustrate that the CEEMDAN-KPCA gets better result in signal de-noising. Compared with different swarm intelligence algorithms, the RUL prediction rates of wind turbine gearbox are improved by using the FOA-ELM prediction model with multiple parameters.
Particle contamination in the dry etching chamber is one of the major issues in the semiconductor manufacturing. Particles on the wafer surface may cause wafer defects and yield loss. Therefore, particle monitoring is important to support contamination control and predictive maintenance activities. To achieve this goal, the present work proposes a virtual metrology (VM) method to estimate the deposit accumulation on the chamber wall, which can be a major root cause of the particle contamination. In the proposed method, the Piecewise Linear Approximation (PLA) is employed and modified to derive a general segmentation template for the trace signals. Based on the trace segmentation template, important features that relate to the deposit accumulation are better extracted and modeled. To justify the effectiveness of the proposed method, we validate our method on the etching data from three different maintenance cycles. In the results, the proposed method with trace segmentation has improved performance compared to the VM model that does not segment the data, as well as the VM model that partitions the trace data by step number. The deposit accumulation indicator given by the proposed method demonstrates a slow trend over etching cycles and it matches well with the off-line metrology measurements.
Sparsity is becoming a more and more important topic in the area of machine learning and
signal processing recently. One big family of sparse measures in current literature is the
generalized lp=lq norm, which is scale invariant and is widely regarded as normalized lp
norm. However, the characteristics of the generalized lp=lq norm are still less discussed
and its application to the condition monitoring of rotating devices has been still unexplored.
In this study, we firstly discuss the characteristics of the generalized lp=lq norm
for sparse optimization and then propose a method of sparse filtering with the generalized
lp=lq norm for the purpose of impulsive signature enhancement. Further driven by the trend
of industrial big data and the need of reducing maintenance cost for industrial equipment,
the proposed sparse filter is customized for vibration signal processing and also implemented
on bearing and gearbox for the purpose of condition monitoring. Based on the
results from the industrial implementations in this paper, the proposed method has been
found to be a promising tool for impulsive feature enhancement, and the superiority of the
proposed method over previous methods is also demonstrated.
Feature extraction plays a vital role in intelligent fault diagnosis of mechanical system. Nevertheless, traditional feature extraction methods suffer from three problems, which are 1) the requirements of domain expertise and prior knowledge, 2) the sensitive to the changes of mechanical system and 3) the limitations of mining new features. It is attractive and meaningful to investigate an automatic feature extraction method, which can adaptively learn features from raw data and discover new fault-sensitive features. Deep learning has been widely used in image analysis and speech recognition with great success. The key advantage of this method lies into the ability of mining representative information and sensitive features from raw data. However, the application of deep learning in feature leaning for mechanical diagnosis is still few, and limited studies have been carried out to compare the effectiveness of feature leaning with various data types. This paper will focus on developing a convolutional neural network (CNN) to learn features directly from frequency data of vibration signals and testing the different performance of feature learning from raw data, frequency spectrum and combined time-frequency data. Manual features from time domain, frequency domain and wavelet domain as well as three common intelligent methods are used as comparisons. The effectiveness of the proposed method is validated through PHM 2009 gearbox challenge data and a planetary gearbox test rig. The results demonstrate that the proposed method is able to learn features adaptively from frequency data and achieve higher diagnosis accuracy than other comparative methods.
Effectiveness of tool condition monitoring strategy depends on accuracy in failure prediction (prognostics) of cutting tools. Data driven approaches are generally used for prognostics of cutting tools. Various prognostics models have been proposed in the literature. Performance of these models in terms of accuracy and applicability are found to be the major constraints for use in real industrial applications. Moreover, application of these models is mainly limited to wear prediction. Extension of such models for remaining life prediction is not explored adequately in the literature. The main contribution of this paper is the development of accurate and applicable data driven models for tool wear estimation and remaining useful life prediction of high speed Computer Numerical Control (CNC) milling machine cutters. These models are developed and validated based on experimental data. Proposed models have demonstrated better results in terms of predicting cutter wear as compared to those mentioned in the literature. It also helps in predicting remaining useful life of cutters under following two industrial cases: - Case I: When only online monitoring data are available. - Case II: When incidental (or planned) offline inspection data are also available.
A simulation model for wear profile prediction of IC engine bearings is presented. Most of the existing studies on clearance joints in multi-body dynamics simulations are based on elementary mechanisms running at constant speed. The present simulation includes parameters and variables necessary for representing real engine kinetics. The developed scheme is capable to simulate an engine running at different conditions, and yields the parameters that are the main contributors to wear. Based on real engine data, the outputs of the simulation were obtained and used to predict the wear depths at locations phase locked to crank angular displacement. The results were compared with wear based results in previous study. The potential of integrating simulated wear and vibration based indicators is also discussed.
We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif-ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make train-ing faster, we used non-saturating neurons and a very efficient GPU implemen-tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.
We report on a series of experiments with convolutional neural networks (CNN)
trained on top of pre-trained word vectors for sentence-level classification
tasks. We first show that a simple CNN with little hyperparameter tuning and
static vectors achieves excellent results on multiple benchmarks. Learning
task-specific vectors through fine-tuning offers further gains in performance.
We additionally propose a simple modification to the architecture to allow for
the use of both task-specific and static word vectors. The CNN models discussed
herein improve upon the state-of-the-art on 4 out of 7 tasks, which include
sentiment analysis and question classification.
Surface roughness is an important measure of quality in metalcutting. Because the surface roughness measure is related to cutting tool wear, this paper develops an efficient surface roughness monitoring and control procedure that combines statistical analysis and the physics of the cutting tool wear mechanism. Examples and data are taken from an actual machining operation.
This paper presents a novel machine performance degradation scheme based on fixed cycle features test (FCFT). Instead of monitoring the machine under constant working load, FCFT introduces a new testing method which obtains data during the transient periods of different working loads. A novel performance assessment method based on those transient data without failure history is proposed. Wavelet packet analysis (WPA) is applied to extract features which capture the dynamic characteristics from the non-stationary vibration data. Principal component analysis (PCA) is used to reduce the dimension of the feature space. Gaussian mixture model (GMM) is utilized to approximate the density distribution of the lower-dimensional feature space which consists of the major principal components. The performance index of the machine is calculated based on the overlap between the distribution of the baseline feature space and that of the testing feature space. Bayesian information criterion (BIC) is used to determine the number of mixtures for the GMM and a density boosting method is applied to achieve better accuracy of the distribution estimation. A case study for a chiller system performance assessment is used as an example to validate the effectiveness of the proposed method.
This paper provides an overview and analysis of statistical process monitoring methods for fault detection, identification and reconstruction. Several fault detection indices in the literature are analyzed and unified. Fault reconstruction for both sensor and process faults is presented which extends the traditional missing value replacement method. Fault diagnosis methods that have appeared recently are reviewed. The reconstruction-based approach and the contribution-based approach are analyzed and compared with simulation and industrial examples. The complementary nature of the reconstruction- and contribution-based approaches is highlighted. An industrial example of polyester film process monitoring is given to demonstrate the power of the contribution- and reconstruction-based approaches in a hierarchical monitoring framework. Finally we demonstrate that the reconstruction-based framework provides a convenient way for fault analysis, including fault detectability, reconstructability and identifiability conditions, resolving many theoretical issues in process monitoring. Additional topics are summarized at the end of the paper for future investigation. Copyright © 2003 John Wiley & Sons, Ltd.
This paper presents an on-line method for detecting damaged teeth in the bandsaw using acoustic emission (AE) signal energy.
The method is based on an analysis of differences in AE energies generated by normal and damaged teeth during sawing. Because
of the difference in the amount of sawing, the AE energy was low for sawing by the damaged tooth and high for sawing by the
normal tooth immediately after the damaged tooth. The ratio of AE energy for two successive teeth — a normal tooth immediately
following a damaged tooth — was much greater than 1, whereas the ratio of AE energy for two successive normal teeth was close
to 1. The results demonstrate that the technique using the AE energy ratio for two successive teeth is effective for on-line
detection of damaged bandsaw teeth.
Traditionally, metal working fluids (MWF) are known to improve machining performance despite poor ecological and health side effects. A new sustainable process that has minimized the use and application of MWFs is NDM (near dry machining). Although there is much controversy on the effectiveness of NDM, it is agreed that a lack of science-based modeling prevents its widespread use. This paper presents a new method to predict tool-wear/tool-life performance in NDM by extending a Taylor speed-based dry machining equation. Experimental work and validation of the model was performed in an automotive production environment in the machining of steel wheel rims. Machining experiments and validation of the new equation reveal that tool-wear can be predicted within 10% when the effect of NDM is statistically different than dry machining. Tool-wear measurements obtained during the validation of the model showed that NDM can improve tool-wear/tool-life over four times compared to dry machining which underlines the need to develop sustainable models to match current practices.
Tool condition monitoring has gained considerable importance in the manufacturing industry over the preceding two decades, as it significantly influences the process economy and the machined part quality. Recent advances in the field of image processing technology have led to the development of various in-cycle vision sensors that can provide a direct and indirect estimate of the tool condition. These sensors are characterised by their measurement flexibility, high spatial resolution and good accuracy. This paper provides a review of the basic principle, the instrumentation and the various processing schemes involved in the development of these sensors.
We present an application of back-propagation networks to handwritten digit recognition. Minimal preprocessing of the data was required, but architecture of the network was highly constrained and specifically designed for the task. The input of the network consists of normalized images of isolated digits. The method has 1% error rate and about a 9% reject rate on zipcode digits provided by the U.S. Postal Service. 1 INTRODUCTION The main point of this paper is to show that large back-propagation (BP) networks can be applied to real image-recognition problems without a large, complex preprocessing stage requiring detailed engineering. Unlike most previous work on the subject (Denker et al., 1989), the learning network is directly fed with images, rather than feature vectors, thus demonstrating the ability of BP networks to deal with large amounts of low level information. Previous work performed on simple digit images (Le Cun, 1989) showed that the architecture of the network s...
Bandsaw machine health monitoring system
P.-Y. H. et al. Huang, Mu-Shui, Jay Lee, Ying-Fan Wu, Hung-Chang Chang, Hung-Lung
Chung, Hsiang Huang, "Bandsaw machine health monitoring system. U. S. Patent
9,901,998," 2018.
Smart Machining Starts with Smart Sawing, Modern Machine Shop
- D Korn
D. Korn, "Smart Machining Starts with Smart Sawing," Modern Machine Shop, 2018.
[Online]. Available: https://www.mmsonline.com/blog/post/smart-machining-starts-withsmartsawing?utm_source=Listrak&utm_medium=Email&utm_term=https%3A%2F%2Fwww.
mmsonline.com%2Fblog%2Fpost%2Fsmart-machining-starts-with-smart-sawing&utm_campaign=MMSExtra&utm_content=MMSExtra.
Statistical surface roughness checking procedure based on a cutting tool wear model Kai Yang and Angus Jeang, pp. 1–8
- Yang